When should I use Iterable vs. Collection in Java?

Question

What are the considerations of using Iterable<T> vs. Collection<T> in Java?

For example, consider implementing a type that is primarily concerned with containing a collection of Foos, and some associated metadata. The constructor of this type allows one-time initialisation of the object list. (The metadata can be set later.) What type should this constructor accept? Iterable<Foo>, or Collection<Foo>?

What are the considerations for this decision?

Following the pattern set forth by library types such as ArrayList (which can be initialised from any Collection, but not an Iterable) would lead me to use Collection<Foo>.

But why not accept Iterable<Foo>, given that this is is sufficient for the initialisation needs? Why demand a higher level of functionality (Collection) from the consumer, than what is strictly necessary (Iterable)?

Answer 1

According to the principle of least surprise, you should emulate the Java collection pattern and take a Collection constructor arg. It will make the people who come after you slightly less puzzled.

Answer 2

In a situation where you may me streaming, like reading a big XML file from a socket or off disk. A Collection doesn't make sense but you could create an iterator to go over the nodes and implement hasNext(), next() and (optionally) remove().

Answer 3

Many of the collection types existed before Iterable<T> (which was only introduced in 1.5) - there was little reason to add a constructor to accept Iterable<T> as well as Collection<T> but changing the existing constructor would have been a breaking change.

Personally I would use Iterable<T> if that allows you to do everything you want it to. It's more flexible for callers, and in particular it lets you do relatively easy filtering/projection/etc using the Google Java Collections (and no doubt similar libraries).

Answer 4

You're correct, as it's considered good practice to ask for the most general form of what you need.

Answer 5

Use the most general interface that you can. Since all you're going to do is iterate, then I would say Iterable is the way to go (since it allows lazy iterators, etc.). You don't care where the iterator is coming from, so don't restrict it more than you have to.

Answer 6

If you go for Collection, then your class can be initialised from a collection only, if you go for Iterable, then you can initialise from either collection or iterable.

As the effort and performance for both is going to be the same, it makes total sense to accept Iterable in the constructor.

Answer 7

An Iterable produces Iterator objects. An Iterator object, by definition, iterates. Notice, that the Iterator interface makes no promise as to how many times next() can be called before hasNext() returns false. An Iterator could possibly iterate over Integer.MAX_VALUE + 1 values before its hasNext() method returns false.

However, a Collection is a special form of Iterable. Because a Collection cannot have more than Integer.MAX_VALUE elements (by virtue of the size() method), it is naturally presumed that its Iterator objects will not iterate over this many elements.

Therefore, by accepting a Collection rather than an Iterable, your class can have some guarantee over how many elements are being passed in. This is especially desirable if your class is itself a Collection.

Just my two cents...

Answer 8

Some constructors, e.g. ArrayList(Collection c), use the toArray() method of Collection for efficiency.

Answer 9

What about synchronization? If you are going to iterate over a collection you need to acquire a lock on it before getting the iterator.

  synchronized(myList) {
      Iterator iter = myList.iterator();
      while (iter.hasNext()) {
          iter.next();
          ...
      }
  }

If you receive the Iterator instead of the Collection you are not sure if your thread owns the object.

Answer 10

See "Why so much emphasis on Iterators and Iterables?" on the Google Collection FAQ for a decent argument for preferring Iterators, especially when dealing with a lot of data. One analogy that might help is to think the difference between forward-only read-only cursors and scrollable cursors.